Performance of bzr update for lightweight checkout

[John Arbash Meinel]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

On 2/4/2011 3:42 PM, Gordon Tyler wrote:
> On 2/4/2011 4:24 PM, Ulrich Mueller wrote:
>>>>>>> On Fri, 04 Feb 2011, Eli Zaretskii wrote:
> 
>>> It was around 50 min for me, but it's a one-time investment. 
>>> Thereafter, each "bzr update" takes much less time, usually around 
>>> 30 sec (assuming you sync at least once a day). So it's worth your 
>>> while.
> 
>> Strangely enough, when using the mirror at lp:emacs, it is only 15
>> min for me, as compared to the 97 min for savannah.
> 
>> Can this be due to different bzr versions? Or is network bandwidth
>> or hardware at savannah too slow for the load they must handle?
> 
> As I recall, for various reasons savannah uses bzr over HTTP, which is a
> significantly less efficient protocol than bzr over SSH, which is how lp
> URLs are resolved.
> 
> Ciao,
> Gordon
> 

They have (in the last few months) allowed bzr+ssh access. The original
report showed:
...
 bzr://bzr.savannah.gnu.org/emacs/trunk/

Which is a raw TCP-to-smart-server connection.  I don't know if there
was a typo in the string or not. I wouldn't be surprised to see terrible
results for doing a lightweight checkout over http. It isn't great over
bzr+ssh, but http has to read whole blocks and then will often throw
that info away by the time you need it again. (I think the internal
cache is about 50MB, but for something the size of emacs with reading
somewhat randomly, I don't expect good cache coherency.)

John
=:->
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.9 (Cygwin)
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org/

iEYEARECAAYFAk1MdXAACgkQJdeBCYSNAAMsfgCdGzABjlnKd16jy7O7+XX7O34T
TBUAn2Dj6i5C9KKu/+ZHbeKC8tJRfeYn
=VQjI
-----END PGP SIGNATURE-----